QUARTERLY JOURNAL OF THE ROYAL METEOROLOGICAL SOCIETY
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
Published online in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/qj.124
Variable cirrus shading during CSIP IOP 5. I: Effects on the
initiation of convection
J. H. Marsham,a * C. J. Morcrette,b K. A. Browning,b A. M. Blyth,a D. J. Parker,a
U. Corsmeier,c N. Kalthoffc and M. Kohlerc
a
University of Leeds, UK
University of Reading, UK
Universität/Forschungszentrum Karlsruhe, Germany
b
c
ABSTRACT: Observations from the Convective Storm Initiation Project (CSIP) show that on 29 June 2005 (Intensive
Observation Period 5) cirrus patches left over from previous thunderstorms reduced surface sensible and latent heat fluxes
in the CSIP area. Large-eddy model (LEM) simulations, using moving positive surface-flux anomalies, show that we expect
the observed moving gaps in the cirrus cover to significantly aid convective initiation. In these simulations, the timing of
the CI is largely determined by the amount of heat added to the boundary layer, but weak convergence at the rear edge of
the moving anomalies is also significant.
Meteosat and rain-radar data are combined to determine the position of convective initiation for all 25 trackable showers
in the CSIP area. The results are consistent with the LEM simulations, with showers initiating at the rear edge of gaps, at
the leading edge of the anvil, or in clear skies, in all but one of the cases. The initiation occurs in relatively clear skies in
all but two of the cases, with the exceptions probably linked to orographic effects.
For numerical weather prediction, the case highlights the importance of predicting and assimilating cloud cover. The
results show that in the absence of stronger forcings, weak forcings, such as from the observed cirrus shading, can
determine the precise location and timing of convective initiation. In such cases, since the effects of shading by cirrus
anvils from previous convective storms are relatively unpredictable, this is expected to limit the predictability of the
convective initiation. Copyright  2007 Royal Meteorological Society
KEY WORDS
cloud shading; nonclassical mesoscale circulation; flux inhomogeneity; convective inhibition
Received 11 December 2006; Revised 24 May 2007; Accepted 31 May 2007
1.
Introduction
Predicting the precise location and timing of convective
storms remains a major challenge for numerical weatherprediction (NWP) models. Although the new generation
of non-hydrostatic models, which use grid spacings of the
order of one kilometre, can sometimes give reasonable
predictions of such storms (e.g. Golding et al., 2005),
they cannot consistently forecast them (Clark and Lean,
2006). The Convective Storm Initiation Project (CSIP)
(Browning et al., 2006a), which took place in the summer
of 2005 in southern England, aims to improve our
understanding of the mechanisms that initiate convective
storms in the mid-latitude maritime climate of the UK.
This understanding can then be used to understand the
likely sources of error in high-resolution NWP forecasts,
and so improve NWP.
This paper explores the effects of moving cirrus anvils
on the convective initiation (CI) observed on 29 June
2005, during CSIP Intensive Observation Period (IOP) 5.
* Correspondence to: J. H. Marsham, Institute for Atmospheric Science, School of Earth and Environment, University of Leeds, Leeds,
LS2 9JT, UK. E-mail: [email protected]
Copyright  2007 Royal Meteorological Society
These anvils were generated by storms over France, and
then advected over the UK. A number of the showers that
developed during IOP 5 did so within the area of southern
England where numerous instruments were deployed as
part of CSIP (Browning et al., 2006a). A subjective
examination of the location of the first appearance of
these showers, made during the course of nowcasting
for CSIP, suggested that the presence of cirrus shields
may have been influencing the locations of CI (Browning
et al., 2006b), despite the cirrus moving at approximately
15 ms−1 .
It is well known that mesoscale spatial variations in
fluxes can induce circulations, such as the sea breezes
that frequently affect CI in the UK (Bennett et al., 2006).
In general, variations in cloud cover, or other parameters
such as soil wetness or vegetation, can also induce such
circulations (called ‘nonclassical mesoscale circulations’
by Segal and Arritt (1992)).
Segal et al. (1986) described idealized two-dimensional
simulations of circulations driven by variable cloud
cover. The contrast between an extensive clear area
(about 80 km across) and an extensive cloudy area was
shown to induce a circulation comparable to a sea breeze,
when the clouds reduced solar fluxes by 60%. They
1644
J. H. MARSHAM ET AL.
noted that even two hours of partial cloud cover can
lead to the generation of a circulation that may be of
meteorological significance; but all model runs were performed with slower-moving cloud-cover variations than
were observed in CSIP IOP 5 (1.7 ms−1 compared with
approximately 15 ms−1 ).
Cases of cloud shading affecting CI had been observed
previously. Bailey et al. (1981) showed that the contrast between two extensive areas of cloudy and clear
sky (about 300 km across) appears to have generated
a mesoscale circulation, which initiated storms; and
McNider et al. (1984) showed that a mesoscale circulation driven by cloud-cover variations was responsible for
the initiation of a line of thunderstorms in Texas. However, both these cases involved more extensive areas of
slower-moving cloud cover than were observed during
CSIP IOP 5. Roebber et al. (2002) described a case from
Oklahoma and Kansas (USA), where a cirrus shield was
important in limiting the development of convection and
reducing the competition between storms. Their results
also suggested that gaps in the moving cirrus cover (diameters of 50 km or more, with speeds of approximately
50 km/h) could be linked to the initiation of convection.
However, this effect on CI was not evaluated statistically,
and the relative contributions of boundary-layer warming
and convergence induced by the gaps were not evaluated.
Section 2 of this paper briefly describes CSIP IOP 5.
Section 3 quantifies the observed effects of the cirrus cover on the surface fluxes. Idealized modelling,
described in Section 4, is then used to investigate what
effects we expect such moving surface-flux anomalies to
have on CI. The observed locations of the CI relative to
the cirrus clouds are then discussed in Section 5, in the
context of these modelling results.
The observed effects of the cirrus shading on the
boundary layer are discussed in the second part of
this study (Marsham et al., 2007). There, we show
that the boundary layer was approximately 0.8 g kg−1
drier under thick cirrus, and that this cirrus led to the
formation of a stable internal layer at the surface, fewer
warm moist updraughts, and in some places a decreased
boundary-layer height. These effects on the boundary
layer are expected to have inhibited CI, thus supporting
the conclusions of this paper.
2. Synoptic situation and overview of convective
initiation in the CSIP area
A mature cyclone was present over the British Isles on 29
June 2005 (Figure 1). The CSIP area, in southern England, was behind (south of) the occluded front associated
with this cyclone. Thunderstorms over northwest France
at 06:00 UTC generated long-lived cirrus anvils. After
the demise of these storms, the orphaned anvils were
advected across the Channel, reaching the CSIP area at
around 09:00 UTC. From around 10:30 UTC onwards,
convective showers initiated over southern England and
quickly developed into a series of thunderstorms. These
storms merged to form a band of heavy precipitation
extending east–west across Britain (Figure 2). As a
result of this heavy rain, some flash-flooding occurred in
Oxfordshire (approximately 100 km north of Chilbolton).
Figure 3 shows the orphaned cirrus anvils. These
appear as blue areas in the satellite images, since they
are cold and reflect solar radiation relatively well. The
rain-radar data show that no precipitation from these
anvils reached the surface. Convective clouds, which
appear as unevenly-coloured yellow regions, were also
Figure 1. Met Office analysis of mean sea-level pressure at 12:00 UTC on 29 June 2005 (CSIP IOP 5).
Copyright  2007 Royal Meteorological Society
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
observed. Some of these produced significant precipitation, as shown by the rain-radar data. Figure 3 shows a
cumulonimbus appearing to form approximately 20 km
northeast of Chilbolton on the edge of a roughlycircular gap (diameter about 25 km) at 12:00 UTC. By
13:00 UTC there were two significant showers in this
gap, which had grown in size and been advected northwards.
1645
3. Observed effects of the cirrus cover on surface
fluxes
Figure 4 shows the radiative effects of the cirrus anvils
at Chilbolton, where solar and surface fluxes were measured. Surface heat fluxes (Figure 4(a)) were measured
by eddy correlation (Andreas et al., 2005; Kalthoff et al.,
2006). Figure 4(b) shows visible-wavelength top-of-
Figure 2. Radar rain rates at 15:00 UTC on 29 June 2005 (CSIP IOP 5). The Chilbolton radar is at the centre of the range rings, shown at 25 km
intervals. Part of the line of storms approximately 100 km north of Chilbolton developed from showers that initiated near Chilbolton, which are
the subject of this study.
Figure 3. Left: false-colour Meteosat images. Visible top-of-atmosphere reflectance is shown in red and green, and 11 µm brightness temperature
is shown in blue, so that cirrus appears blue and cumulus yellow. Positions of surface-flux observations are indicated by white crosses (with
Bath west of Chilbolton). Right: rain-radar data (the colour scale is the same as in Figure 2). Both Meteosat and radar plots are shown at 12:00
and 13:00 UTC, and range rings are centred on the Chilbolton radar and shown at 25 km intervals. The apparent difference between the satellite
and radar positions of the easternmost storm at 13:00 UTC is due to parallax.
Copyright  2007 Royal Meteorological Society
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1646
J. H. MARSHAM ET AL.
Figure 4. Time series of observations from Chilbolton (with the symbol/line and sampling period shown in brackets): (a) downwelling solar heat
flux (solid line, 10 s); downwelling infrared flux (dotted line, 10 s); upward surface sensible and latent heat fluxes (dashed and dash-dotted lines
respectively, 15 min). (b) Meteosat observations from the pixel corresponding to Chilbolton: visible TOAR (dashed line, 15 min); infrared BT
(solid line, 15 min, with a 240 K offset); liquid-water path (dash-dotted line, 10 s) retrieved from an upwards-pointing dual-wavelength (22 GHz
and 28 GHz) microwave radiometer at Chilbolton (Simpson et al., 2002). (c) 2 m air temperature (dashed line, 10 s); boundary-layer potential
temperatures from radiosondes averaged over lowest 400 m (crosses, 1 h or 2 h).
atmosphere reflectance (TOAR) and brightness temperature (BT) from Meteosat-8: cirrus clouds give TOARs
of 30% or more and low 11 µm BTs, while water clouds
give high TOARs and higher BTs.
Figures 4 (a) and (b) show that, between approximately
10:00 UTC and 13:00 UTC, cirrus clouds significantly
reduced the incoming solar radiation at Chilbolton, and
consequently reduced the upwards sensible heat flux
measured at the surface (although Figure 4(b) shows that
liquid water added to the effect of the cirrus to give
the minimum in observed fluxes at around 12:00 UTC).
The cirrus also increased the downwelling infrared flux,
although this effect is negligible compared with the
effect on the downwelling solar flux. The decrease in
surface sensible heat flux caused by the cirrus coverage
reduced the strength of the super-adiabatic surface layer,
and as a result reduced the difference between the 2 m
air temperature and the boundary-layer air temperature
(Figure 4(c)). Similar effects were observed at Bath
(approximately 65 km west of Chilbolton, as shown in
Figure 3), where the cirrus reduced the incoming solar
fluxes between about 11:00 and 13:30 UTC (not shown).
The surface-flux variations associated with the cirrus
in the CSIP area can be estimated using the observed
linear correlations between the downwelling solar heat
Copyright  2007 Royal Meteorological Society
flux measured at the surface and the surface sensible and
latent heat fluxes (Figure 5). By extrapolation from the
data shown in Figure 5, at Chilbolton at 12:00 UTC the
clear-sky solar irradiance of approximately 1100 Wm−2
that would have been experienced in the absence of
cloud would have given a sensible heat flux of approximately 210 Wm−2 . In reality, at 12:00 UTC cirrus and
water clouds reduced the solar flux to approximately
200 Wm−2 , and hence the sensible flux to approximately zero. At around 11:00 UTC, cirrus cover reduced
the solar flux to approximately 400 Wm−2 , and hence
the sensible flux to approximately 50 Wm−2 . Latent
heat fluxes varied linearly with sensible fluxes at both
Chilbolton and Bath, and at both sites a zero sensible
heat flux still corresponded to a positive latent heat flux
(Figure 5). So, given the spatial distribution in downwelling solar radiation, we can estimate spatial distributions of surface sensible flux in the CSIP area, assuming
that the surface behaves similarly to that of the fluxmeasurement sites at Chilbolton or Bath (i.e. flat grass).
Meteosat data were used to estimate the spatial variations in downwelling solar fluxes, and hence to infer
surface fluxes. Downwelling solar radiation was measured at the surface at Chilbolton and Bath, as well as
on the Dornier 128 aircraft, which was flying at a height
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
1647
Figure 5. Observed surface and solar fluxes at (a) Chilbolton and (b) Bath, from 10:00 to 14:00 UTC. Solid lines show least-squares fits to data
from the named site. Dashed lines show fits to data from the other site.
of about 500 m from 09:00 to 12:38 UTC. The expected
clear-sky downwelling solar flux was calculated at hourly
intervals using the observed profile and the Fu–Liou radiation code (Fu and Liou, 1992, 1993), and these data were
then interpolated to give a value for each observation
time.
Figure 6 shows the correlations between the solar
transmission (i.e. the observed downwelling solar radiation divided by the calculated clear-sky value) and the
Meteosat data, for the visible TOARs and the 11 µm BTs.
(Observed errors in the geolocation of the Meteosat data
were corrected, and a parallax correction was applied to
all pixels (Johnson et al., 1994), using a cloud-top height
of 6 km. This height was derived from vertical scans by
the high-power radars at Chilbolton. The Meteosat pixel
used was then shifted by up to one pixel north or south
to give the best correlations, to allow for uncertainties in
these two corrections.) The linear fits calculated for the
two sites were similar, except where an outlying observation with a high TOAR gave a shallower gradient for
the Bath data.
Figure 7 shows surface sensible heat fluxes calculated from Meteosat data, using the linear fits between
Meteosat data and solar transmission (Figure 6), and
between surface sensible heat flux and downwelling
solar radiation (Figure 5), observed at Chilbolton. The
Chilbolton site was chosen, rather than Bath, because the
precipitating storms studied were nearer to Chilbolton.
These estimated fluxes are valid only over land and
for surfaces similar to those where the eddy-correlation
fluxes were obtained (i.e. over flat grass). The fluxes
are also only valid for cloud conditions similar to those
observed. Thus, although Figures 7 (a) and (b) show
broadly similar patterns, they do show significantly different fluxes in some areas. Both show extensive areas
outside the cirrus coverage (Figure 3) with surface sensible fluxes of approximately 200 Wm−2 , values of
0–50 Wm−2 under the cirrus, and gaps in the cirrus cover
with fluxes of about 150 Wm−2 . However, significant
differences between Figures 7 (a) and (b) do occur, for
example 100 km north of Chilbolton, in an area affected
Copyright  2007 Royal Meteorological Society
by low stratiform cloud, and 75 km west-southwest of
Chilbolton, in an area with widespread shallow cumulus.
These occur because the correlations between Meteosat
observations and solar transmission were derived largely
for cirrus cover. However, coverage of low stratiform
cloud was very limited in the area of interest during the
hours preceding CI, and we show in Section 4.5 that the
effects on modelled CI of neglecting the shading from,
and radiative heating of, developing cumulus are relatively small.
It is of interest to compare the magnitude of the
variations in surface fluxes induced by the moving cirrus
anvils with typical effects of land-use variations on
surface fluxes. In southern England, apart from land/sea
effects, the largest land-use effects are from urban versus
rural areas. Urban areas are expected to increase daytime
surface sensible fluxes by up to a factor of two and
decrease latent heat fluxes by a factor of between one
and six (Thielen et al., 2000). There are, however, no
major urban areas in the CSIP area (the largest urban
areas have diameters of approximately 15 km). Variations
in vegetation are likely to have effects of up to only
a factor of 2.5 (Segal and Arritt, 1992), and vegetation
variations in southern England also tend to occur on small
spatial scales, so have limited potential for significant
development of mesoscale circulations. In addition, urban
areas and vegetation areas essentially alter only the
Bowen ratio, which does not alter the flux of equivalent
potential temperature (Betts et al., 1996), while gaps in
the cirrus shading increase both sensible and latent fluxes,
and so increase the equivalent-potential-temperature flux.
However, variations in Bowen ratio, due to variations
in surface properties, can result in differences in the
dependence of the total surface heat flux (sensible plus
latent) on boundary-layer cloud cover. This effect has
been neglected; but, as discussed above, the vegetation
cover of southern England is relatively uniform.
In conclusion, the magnitudes of the cloud-induced
variations in fluxes in this case are probably greater than
any urban or vegetation effects. However, in contrast
with land-use variations, the observed cloud-induced
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1648
J. H. MARSHAM ET AL.
Figure 6. Plots of estimated transmission of solar radiation and Meteosat visible TOAR (a, c, e) and 11 µm BT (b, d, f). Surface data from
Chilbolton (a, b) and Bath (c, d) are shown, as well as data from the Dornier 128 aircraft (e, f). For surface data (a, b, c, d), transmission is
mean transmission over 200 s (visible) and 600 s (infrared) (i.e. the time taken for a cloud moving at 5 ms−1 to traverse a Meteosat pixel of
1 km or 3 km). Error bars show the standard deviation in the Meteosat observations over 3 × 3 pixels, and the standard deviation in the solar
transmission over 400 s (visible) and 1200 s (infrared) (10 s-resolution data from Chilbolton and 10 min-resolution data from Bath). For aircraft
data (e, f), transmission is mean transmission over 10 s (visible) and 30 s (infrared) (i.e. approximately the time taken for the aircraft to traverse
a Meteosat pixel of 1 km or 3 km). Straight-line fits are for Chilbolton (solid line), Bath (dashed line), and the aircraft data (dotted line).
variations are moving relative to the surface, and so the
effects are short-lived for a given area. The effects of this
motion are investigated in the next section.
4. Modelling the effects of moving surface-flux
anomalies
Despite the significant approximations in the estimation
of effects of the cirrus cloud on the surface fluxes
(Figure 7), the estimated fluxes are sufficiently precise
for setting up idealized modelling studies, using the
Met Office large-eddy model (LEM) (Gray et al., 2001;
e.g. Grabowski et al., 2006). These modelling studies test
the sensitivity of modelled CI to flux variations of similar
magnitudes to those observed during CSIP IOP 5.
Copyright  2007 Royal Meteorological Society
The various model runs used are summarized in
Table I. Runs using moving positive surface-flux anomalies (POS2D and POS3D) were used to investigate the role of
gaps in the cirrus cover in CI. Runs using a moving negative surface-flux anomaly (NEG2D) were used to investigate the development of convection along the outer edges
of the cirrus anvil. Finally, runs with a radiative-transfer
scheme coupled to the LEM (FULLRADN) were used to
investigate the effects of neglecting the shading by convective clouds and the radiative heating of those clouds.
4.1.
Method
Simulations using version 2.3 of the LEM were used
to simulate the effects of moving surface-flux anomalies. The LEM is a non-hydrostatic model that can be
run in one, two or three dimensions. Although the LEM
can be used with a three-phase microphysics scheme,
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1649
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
Figure 7. Estimated surface sensible heat flux at 12:00 UTC, calculated using the observed correlation between solar and sensible fluxes (Figure 5),
and the mean correlation between solar transmission and (a) Meteosat visible TOAR and (b) Meteosat 11 µm BT (Figure 6). These fluxes were
derived using surface-flux observations from two locations, and so errors are large in some areas; fluxes are not valid over the sea.
Table I. Summary of sets of model runs used in this paper.
Name
POS2D
POS3D
NEG2D
FULLRADN
Domain size
500
500
500
500
km
km by 3 × D
km
km
Unperturbed fluxes
(sensible & latent, Wm−2 )
60 & 180
60 & 180
240 & 480
f (solar): see Figure 5
only cloud-water and rain (a Kessler rain scheme)
were included, since the process of CI that was being
investigated occurred before the ice-phase processes
were significant (the melting level was at approximately
3000 m, about 1400 m above the level of free convection). The model domain was 20 km deep, and Rayleigh
damping was applied above 13 km. For two-dimensional
simulations, we used horizontal grid spacings of 200 m
and vertical grid spacings of 50–100 m in the boundary layer and 100 m up to heights of approximately
5000 m, with progressively larger vertical grid spacings
above 5000 m. Horizontal grid spacings of 1 km were
used in three-dimensional simulations (POS3D), because
of computational constraints, and vertical grid spacings
were identical to those used in two-dimensional simulations. Periodic lateral boundary conditions were used in
all simulations.
The model was initialized with the radiosonde
sounding, shown in Figure 8, made from Reading at
11:15 UTC. Reading, approximately 50 km northeast of
Chilbolton, was the closest site to the initiation of the
deeper cells observed northeast of Chilbolton between
12:00 and 13:00 UTC. For two-dimensional simulations,
the model domain was aligned with the observed wind at
the level of the cirrus clouds. This gave boundary-layer
winds of up to 5 ms−1 (Figure 8(b)).
Copyright  2007 Royal Meteorological Society
M
D
(km)
v
(ms−1 )
2 to 5
5
0.25 to 0.5
0.5
2 to 50
30
100
30
0 to 15
10 and 15
0 to 15
10
In all runs, it was assumed that the only effect of the
cirrus was to modify the downwelling solar radiation at
the surface, and hence the surface fluxes; this is justified,
because vertical scans by the Chilbolton radars show
that the cirrus did not extend much below 5 km and
falling ice could not reach convective storms initiating at
around 1600 m (the level of free convection). Radiative
heating of the cloud and atmosphere, as well as shading
of the surface by the developing convection, were also
neglected in most simulations (POS2D, POS3D and NEG2D).
These last assumptions were tested in a third set of
simulations (FULLRADN), which used a radiative-transfer
scheme coupled to the LEM.
For runs used to investigate the role of gaps in the
cirrus cover in CI (POS2D and POS3D), uniform sensible
and latent fluxes of 60 Wm−2 and 180 Wm−2 respectively were applied at the surface in the model, except in
the moving flux anomaly, where fluxes were multiplied
by a factor M. This positive flux anomaly represented
the effect of a gap in the cirrus cover. The width of the
flux anomaly, D, was varied between 2 km and 50 km.
This range of scales encompasses the 25 km size of
the gap observed northeast of Chilbolton at 12:00 UTC
(Figure 3). The speed of the anomaly was varied between
zero and 15 ms−1 ; the speed of travel of the anvil was
tracked by Meteosat (this is consistent with radiosonde
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1650
J. H. MARSHAM ET AL.
Figure 8. 11:15 UTC radiosonde profile from Reading, one of the nearest radiosonde sites to the early shower-initiation events: (a) tephigram;
(b) wind velocities. In the legend, ‘2D wind’ refers to the wind profile used in two-dimensional LEM simulations (this is the wind velocity in
the direction of the wind velocity at the cirrus altitude).
wind data, which showed wind speeds of approximately
15 ms−1 at the cirrus altitude of approximately 7 km, or
400 hPa). The factor M was varied between 2 and 5.
In the two-dimensional simulations, a large domain of
500 km was used. Thus, an anomaly of width 50 km
moving at 15 ms−1 did not reach the far end of the
domain in a 7 h simulation. In three-dimensional simulations (POS3D), the domain width was three times the
anomaly diameter D (in the dimension perpendicular to
the cirrus steering velocity), and again 500 km long.
To investigate the development of cumulus along the
outer edges of the overall cirrus anvil (NEG2D), uniform
sensible and latent fluxes of 240 Wm−2 and 480 Wm−2
respectively were applied at the surface in the model,
except in the moving flux anomaly where fluxes were
reduced by a factor M. The diameter of the cold anomaly
was fixed at 100 km, approximately the diameter of the
anvil shown in Figure 3, and its speed again varied
between zero and 15 ms−1 . The model domain was
600 km long. Thus, these experiments were very similar
to the positive-anomaly runs, except that in this case the
positive area (i.e. the area not affected by the negative
anomaly) was wide (500 km), whereas the warm area in
the runs with a positive anomaly was narrow (2–50 km).
For the third set of runs (FULLRADN), used to test
the assumption that shading by convective clouds and
radiative heating of convective clouds could be neglected,
a constant top-of-atmosphere downwelling solar flux was
used, calculated for 12:00 UTC at Chilbolton on 29 June
2005. This downwelling flux was halved, except in a
30 km gap moving at 10 ms−1 , which represented the
gap in the cirrus cloud cover. The surface fluxes were
Copyright  2007 Royal Meteorological Society
specified to depend on the downwelling solar radiation
at the surface, using the linear correlation shown for
Chilbolton in Figure 5, allowing shading by convective
clouds to be accounted for. The four-stream radiation
model, with six solar and twelve infrared bands, of (Fu
and Liou, 1992, 1993) was used, with an independent
pixel approximation.
4.2.
Scale analysis
With spatially-uniform and time-invariant surface fluxes,
the time taken for cloud top to reach a particular height
between 1100 m and 3000 m was found to be, to a good
approximation, inversely proportional to the surface flux
(not shown). So, with a surface flux F , the time taken in
the absence of a flux anomaly is
t=
Q
,
F
(1)
where Q is the amount of heat required per unit area. A
moving surface-flux anomaly, with diameter D and speed
v, increasing surface fluxes by a factor M, will affect any
point for a time less than or equal to D/v. Therefore, If
heat is not mixed horizontally, the time taken for the
cloud top to reach a particular height will be reduced by
(M − 1)D/v, if this time depends only on the maximum
amount of heat added to the boundary layer at any point.
The situation is not one-dimensional, though. Figure 9
shows a schematic of the heat added to the boundary layer, if the heat is not mixed horizontally, at time
t > D/v. Using this simple model, we expect the pressure perturbation p induced by the flux anomaly to
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1651
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
be proportional to Q2 − Q1 (Figure 9). Thus, using the
Boussinesq approximation, the maximum pressure gradient induced, dp/dx, is approximated by
g Q2 − Q1
p
≈
,
x
Cp T
D
(2)
where g is the acceleration due to gravity, Cp is the
specific heat capacity of the air, and T is the mean
temperature in the boundary layer. Substituting for Q1
and Q2 (Figure 9) gives:
g (M − 1)F
p
≈
.
x
Cp T
v
(3)
These expressions are equally valid for a positive
or a negative anomaly, and the pressure gradient is
independent of the anomaly diameter D. From the
momentum equation, using
u ≈
1 p
t
,
ρ x
in the three-dimensional case.
This uplift will lift any lid at the top of the boundary
layer, reducing the convective inhibition (CIN):
D
,
v
dCI N
R
=
dt
z0
CI N ≈
(6)
where w is the vertical velocity and H the depth of
the boundary layer. We expect this convergence to be
Q2
Wind
t=0
D
F
F
t=t
D
p0
p
− R Cp
w
dθv
dz,
dz
(8)
− R
Cp
RgH 2 dθv (M − 1)F D
.
z0 ρCp T dz
v3
(9)
RgH 3 dθv (M − 1)D
,
z0 Cp T dz
Mv 3
(10)
if the whole boundary-layer depth has to be heated to
overcome the CIN.
Figure 10 shows another schematic of the heat added
to the boundary layer, this time with t < D/v, again
assuming that heat does not mix horizontally. In this case
the maximum pressure gradient induced is given by:
or:
p
g Q2 − Q1
≈
,
x
Cp T
tv
(11)
p
g (M − 1)F
≈
.
x
Cp T
v
(12)
F
Distance
Figure 9. Schematic of the model set-up for the POS2D and POS3D runs,
with a positive flux anomaly (diameter D, speed v) at time t > D/v.
The initial anomaly position is shown in grey, and the final position
in black. A quantity of heat Q1 has been added to columns unaffected
by the flux anomaly, while for columns over which the whole anomaly
has moved, this is increased to Q2 .
Copyright  2007 Royal Meteorological Society
p0
p
t ≈
MF
F
Thus, with a flux of MF over the anomaly, and a
constant pressure p0 , the time taken for CI is reduced by
Model cross section at t > D/v
Q1=tF
Q2=(t+(M–1)D/v)F
Anomaly speed, v
Q1
(7)
(5)
g (M − 1)F
w
≈
,
H
ρCp T
v2
Total
heat
added
RT dz,
where dθ/dz is the dry adiabatic lapse rate, and where we
have assumed that pressure is constant over the region
of CIN. Over the time (D/v) for which the anomaly
increases the surface fluxes at a point, this will reduce
the CIN by approximately
Hence, from continuity and convergence in two dimensions,
dw
du
=− ,
dz
dx
we get:
where T is the difference in density temperature
between the parcel and its surroundings, R is the gas
constant for dry air, and z0 is the scale height of the
atmosphere. Assuming that the convergence leads to
adiabatic uplift of air at the top of the boundary layer,
this gives:
Equation (3) gives:
g (M − 1)F
u
.
≈
x
ρCp T
v2
1
z0
CI N =
(4)
(where ρ is the air density and u is the velocity
perturbation in time t), and
t =
increased for a circular anomaly in three dimensions,
using
du dv
dw
=−
−
,
dz
dx
dy
Using Equation (4), and
x = vt,
we obtain:
g (M − 1)F
u
≈
.
x
ρCp T
v2
(13)
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1652
J. H. MARSHAM ET AL.
Total
heat
added
Model cross section at t < D/v
Q1=MtF
Q2=tF
Anomaly speed, v
Q2
Q1
Wind
MF
F
t=t
F
F
t=0
tv
tv
Distance
Figure 10. As Figure 9, but for t < D/v and for a negative flux
anomaly (NEG2D).
This is the same expression as Equation (5), for t >
D/v, and again it is valid for positive or negative flux
anomalies.
4.3. Modelled effects of moving positive surface-flux
anomalies
All runs described in this section are two-dimensional
except where stated otherwise. In runs using small
(2–50 km-width) positive flux anomalies (i.e. POS2D and
POS3D in Table I), the modelled cloud-top height tended
to remain at approximately 1 km for some time, and
then grew rapidly through the level of free convection
at 1600 m to approximately 4 km.
For cases where the time taken for the cloud-top to
reach a particular height (the ‘onset time’) is greater than
the transit time (D/v), if the one-dimensional argument
presented in Section 4.2 holds, then the data shown in
Figure 11 should follow straight lines (as indicated by the
diagonal dashed lines). The data in Figure 11 are close
to straight lines, but with slightly steeper gradients than
the 1 : 1 dashed lines shown. This shows that it is largely
the quantity of heat added to any point that determines
when the cloud top reaches a particular height (the onedimensional assumption), but the convergence produced
by the moving anomalies also significantly aids the development of the convection. The data shown in Figures
11 (b) and (c) are further from the dashed straight lines
shown than the data shown in Figure 11(a). This shows
that convergence effects are more significant for the time
taken for the cloud top to reach 1600 m or 3000 m than
1100 m. The level of free convection is approximately
1600 m, so Figure 11(b) shows the timing of CI. There
is a discontinuity in the data shown in Figure 11(b), with
CI approximately 1 h faster for (M − 1)D/v > 0.7. This
does not seem to correspond to any threshold in the quantities D, D/v, M and so on, but suggests that there is
a threshold in (M − 1)D/v for the anomalies to induce
sufficient convergence to affect the modelled CI.
As (Petch, 2006) leads us to expect, increasing the
grid spacing from 200 m to 1 km in the Met Office LEM
increases the time taken for the cloud top to reach its
level of free convection (Figures 11 (b) and (d)). Petch
Figure 11. Time taken for cloud-top height to reach (a) 1100 m, (b, d) 1600 m and (c) 3000 m (‘onset time’), plotted against (M − 1)D/v (the
effect of the positive flux anomaly expected from a one-dimensional argument, Section 4.2), for multiple model runs. Panels (a, b, c) show results
from two-dimensional runs with 200 m grid spacings. Panel (d) shows results from two- and three-dimensional runs with 1 km grid spacings.
Dashed diagonal lines show the results expected from the one-dimensional argument (Section 4.2), with the dashed diagonal line in (d) the same
as that shown in (b). Symbols indicate the magnitude of the flux anomaly. Data within a square are for v = 15 ms−1 and varying D; otherwise
data are for D = 30 km and varying v. Large diamonds indicate where the transit time (D/v) is greater than the onset time. Dotted vertical
lines demarcate the estimated range of observed (M − 1)D/v.
Copyright  2007 Royal Meteorological Society
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
1653
Figure 12. Model cross sections at 3 h (before CI occurs), for a run with a 30 km-diameter positive flux anomaly (M = 4) moving at 10 ms−1 .
The anomaly magnitude was increased from M = 1 to M = 4 over the first 2 h of the simulation. (a) Surface pressure is shown by the solid line.
The vertical dash-dotted lines show the initial and final positions of the positive surface-flux anomaly. The dash-dotted line from left to right
schematically indicates the heat added to each column. (b) As (a), but the solid line shows the horizontally-smoothed vertical velocity averaged
over the boundary-layer depth.
(2006) also shows that for simulations with spatiallyuniform surface fluxes the development of convection
tends to be slower in three dimensions than in two
dimensions. However, in this case the use of three
dimensions decreases the time required for CI, compared
with the two-dimensional case (Figure 11(d)), and this is
likely to be from the increased effects of convergence in
three dimensions (Section 4.2).
The observations suggest that gaps in the cirrus with diameters of 20–30 km and moving at
10–15 ms−1 increased surface fluxes from 50 Wm−2
to 150–200 Wm−2 (Figure 7). Using these ranges of
observed values gives:
0.74 h <
(M − 1)D
< 2.5 h,
v
(as indicated by the vertical dotted lines in Figure 11(b)).
In the two-dimensional LEM simulations, this allowed
CI after approximately 3.5–6 h, rather than the 8 h
associated with a very weak flux anomaly. Thus the LEM
results show that the gaps in the cirrus cover observed
during IOP 5 are expected to have significantly aided CI.
Figure 12(a) shows that the pressure perturbation at
any point reflects the amount of heat added to the column.
For a moving flux anomaly, the pressure perturbation
slowly returns towards the unperturbed value behind the
anomaly. The minimum pressure is located at the rear
edge of the anomaly (Figure 12(a)), and convergence
occurs there. This results in uplift in the boundary layer
at, or just behind, the rear edge of the moving anomaly,
and subsidence at its leading edge (Figure 12(b)).
The position of CI can be understood in terms of
the amount of heat added to the boundary layer and
the location of the boundary-layer convergence. For a
stationary positive flux anomaly and no mean wind,
the convection was initiated directly over the positive
flux anomaly, as expected. With a mean wind, the
convection was initiated towards the downwind side
of the anomaly (not shown). If the flux anomaly is
moving at 10–15 ms−1 , the clouds tend to deepen at the
rear edge of the anomaly or behind it (e.g. Figure 13).
Copyright  2007 Royal Meteorological Society
Figure 13 also shows that there is a minimum in the
modelled boundary-layer depth over the anomaly, and
the boundary layer at the rear edge of the anomaly
is approximately 200–300 m deeper than in the areas
downwind (i.e. larger x) that are so far unaffected by the
anomaly. Subsidence at the leading edge of the anomaly
leads to the lowered boundary-layer depth observed
over the anomaly, and the increased heat fluxes and
convergence from the anomaly lead to the increased
boundary-layer depth at its rear edge (Figure 12); as
a result, the modelled CI occurs there. If the anomaly
moves at approximately the same speed as the low-level
wind, then the convergence is towards the centre of the
anomaly, and the clouds deepen here first. With no wind
at low levels, the CI again occurs towards the rear edge
of the moving flux anomaly (not shown).
Figure 14(a) shows that the pressure gradient induced
across the positive flux anomaly is approximately proportional to (M − 1)/v, consistent with Equation (3), but
Figure 13. Hovmöller (distance–time) plot of smoothed boundary-layer
depth (shaded, 100 m contour interval) and smoothed cloud-top height
(contours at 1100 m, 1600 m and 3000 m), using a positive surface-flux
anomaly moving at 15 ms−1 . Dashed lines show the track of the
positive flux anomaly. The anomaly magnitude was increased from
M = 1 to M = 4 over the 2 h spin-up period. Dotted lines show
estimated tracks of boundary-layer features formed at the leading edge
of the anomaly at the start and end of the spin-up period.
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1654
J. H. MARSHAM ET AL.
Figure 14. Modelled (a) pressure gradient as a function of (M − 1)/v (Equation (3)), and (b) maximum vertical velocity integrated over the
boundary-layer depth (smoothed over 2 km) as a function of (M − 1)/v 2 (Equation (6)) at 4 h, for runs with no saturation.
the values are approximately three-quarters of those predicted. Figure 14(b) shows the maximum vertical wind
in the boundary layer at the rear edge of the positive
flux anomaly, which is proportional to (M − 1)/v 2 , consistent with Equation (6), but with values approximately
one-third of those predicted. Considering the idealized
nature of the arguments used to derive the theoretical
values, these differences between the modelled and predicted quantities shown in Figure 14 are not surprising.
Figure 15 shows the difference between the model
results and the linear fit shown in Figure 11, as a function of (M − 1)D/Mv 3 (Equation (10)), representing the
impact on CIN. The results correlate to some extent, providing some support for the effects of the convergencedriven uplift outlined in Section 4.2. Again, the scatter
and the differences in the absolute values are certainly
not surprising, given the idealized nature of the scaling
arguments involved. The relationship is not as good as
that between the deviation and (M − 1)D/v, however
(Figure 11(b)); thus this deviation depends more on the
extra heat added to any column by the flux anomaly than
on the estimated impact on CIN (Equation (10)). This
shows that although the theory outlined in Section 4.2
predicts the pressure gradients and uplift well, the effects
on CIN and CI are more complex.
Figure 15. Deviation of data shown in Figure 11(b) from the
straight-line fit, as a function of (M − 1)D/Mv 3 (compare with Equation (10)).
Copyright  2007 Royal Meteorological Society
4.4. Modelled effects of moving negative surface-flux
anomalies
A wide (100 km-diameter) moving negative flux anomaly
was applied to represent the effects of the overall cirrus
anvil (NEG2D runs, Table I), rather than the positive
anomalies used to represent the gaps in Section 4.3. These
simulations are essentially similar to those described
in (Segal et al., 1986), except that the speed of the
cloudy region is much larger here (up to 15 ms−1
compared with 1.7 ms−1 ). Results from (Segal et al.,
1986) suggest that convergence and uplift should be
generated at the edges of the cold anomaly, and that
this uplift should be maintained or enhanced at the
leading edge of the anomaly, but reduced on its trailing
edge.
Two sets of simulations were performed. The first
included no saturation, so vertical velocities were due to
dry convection influenced by the moving cold anomaly.
This allowed the effects of the cold anomaly, once
it had moved far from its starting position, to be
understood without the added complexities of moist
convection. In the second, warm phase, microphysics
were included. This resulted in clouds forming while
the position of the cold anomaly overlapped its initial
position.
Figure 16 shows that without saturation in the model
the moving cold anomaly suppressed convection in its
path, as expected, but there is uplift just inside its
leading edge and subsidence at its trailing edge. This
uplift is reduced when the anomaly speed is increased
(Figure 16(b)), and the uplift and corresponding horizontal pressure gradients are again predicted reasonably well
by the theory described in Section 4.1 (not shown).
With saturation included in the model, clouds form and
deepen in the region of maximum heating, i.e. ahead of
and behind the track of the cold anomaly (Figure 17).
Figure 17(a) shows that with a mean wind, which allows
clouds to move more ‘in phase’ with the anomaly, deeper
convection persists in the region of convergence along
the leading edge of the anomaly. Clouds are suppressed
in the subsidence at the trailing edge of the moving
anomaly, with or without the mean wind in the model
(Figure 17).
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
1655
Figure 16. Vertical velocity w, averaged over the boundary-layer depth after 4 h (so that t > D/v), in model runs without saturation and liquid
water. The thin and thick lines shows data smoothed over 4 km and 40 km, respectively. Results are from a dry run (i.e. with no saturation),
so clouds do not affect the velocities. The observed wind was used. Dotted (dashed) lines give the starting (final) position of the cold anomaly,
which is moving at 15 ms−1 . Anomaly speeds of (a) 10 ms−1 and (b) 15 ms−1 were used.
Figure 17. Hovmöller (distance–time) plots of modelled LWP using a cold anomaly, moving at 15 ms−1 , using (a) observed wind and (b) no
wind. Contours are at 0.01 kg m−2 (dotted lines), 0.1 kg m−2 (thin solid lines) and 1.0 kg m−2 (thick solid lines). Dashed lines show the path
of the moving cold anomaly.
Overall, the suppression of convection by the cold
anomaly is clear, and the results predict enhancement
of convection at the leading edge of the cirrus anvils
during CSIP IOP 5 and subsidence at their trailing edge.
This suggests that the cumulus convection observed at
the leading edge of the anvil (Section 5) may have been
enhanced by circulations induced by the cirrus-cloud
shading.
4.5. Results from simulations using a radiative-transfer
model
Here we test the assumptions used so far: that the effects
of shading by convective clouds and radiative heating
of convective clouds can be neglected. Figure 18 shows
a space–time diagram of cloud-top height from model
runs (FULLRADN in Table I) neglecting and including
these effects. It can be seen that the convection is
fairly insensitive to allowing the convective clouds to
interact with the downwelling radiation. With and without
the fully-coupled radiation, convection develops along
the rear (upwind) edge of the positive flux anomaly
and behind the anomaly (Figure 18). This justifies the
assumption, used in Sections 4.3 and 4.4, that the effects
of the cirrus clouds can be represented by moving
surface-flux anomalies.
Copyright  2007 Royal Meteorological Society
The shading by developing convective clouds reduced
the surface sensible heat fluxes by around 20 Wm−2 .
However, radiative heating of those clouds to some
extent compensated for the reduction in surface fluxes by
increasing the column-integrated radiative heating rates
by up to 10 Wm−2 . These values are small compared
with the magnitude of the effects of the solar flux
anomaly itself, which in these runs increases surfacesensible-heat fluxes from approximately 40 Wm−2 to
125 Wm−2 . As a result, a run allowing for shading by
convective clouds and radiative heating of those clouds
gives very similar CI to a run neglecting both effects
(Figure 18).
5. Analysis of observed convective initiation in the
CSIP area
Section 2 shows the kind of observational data that
led Browning et al. (2006a) to hypothesize that variable
shadowing by moving cirrus anvils may have affected
CI during CSIP IOP 5. Modelling results discussed in
Section 4 show that this is indeed to be expected. In
this section, we examine the observational data in detail,
to see whether these findings are fully substantiated.
In fact, the observed fields of cirrus cloud cover and
convective development were complex, and so the results,
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1656
J. H. MARSHAM ET AL.
Figure 18. Hovmöller (distance–time) plots of smoothed cloud-top heights above 1600 m, (a) without radiative heating of clouds or shading by
convective clouds, and (b) with fully-coupled radiation. The diagonal dashed lines show the path of the moving positive solar flux anomaly,
which increased the modelled downwelling solar flux by a factor of two.
while supporting the hypothesis, are not in themselves
definitive.
Tracking the observed precipitation and the corresponding cumulus convection, using the rain-radar network and Meteosat data respectively, allowed the locations of convective initiation and development to be
determined and compared with the cirrus cloud cover.
The starting point of this analysis was to determine the
trajectories of all precipitation echoes forming within
the CSIP area (a 200 km-by-200 km square centred on
Chilbolton) that could be tracked for at least two frames
of the 15 min data from the radar network. Those echoes
that formed on the outer edge of pre-existing showers,
which are likely to have been due to secondary initiation,
were excluded from the analysis. A total of 25 showers
were tracked in this way.
The high-resolution visible data from Meteosat were
then used to track the specific convective clouds associated with these 25 showers. A parallax correction was
applied to the Meteosat data (Johnson et al., 1994). This
assumed a cloud-top height of 5 km, since radiosonde
data collected in the CSIP area showed that, provided
the lid at around 900 hPa could be overcome, convection from the surface could rise to a height of
around 550 hPa (approximately 5 km). This was also
observed in the LEM simulations. Figure 19 shows that
there is good agreement between the location of the
radar and satellite tracks (black and grey lines, respectively), except that convective clouds were normally
visible before precipitation was detected. This shows
that the parallax correction used is sufficiently accurate for the purpose intended, i.e. determining the location of the convective clouds with respect to the cirrus
clouds.
The 11 µm BT from Meteosat was used to define
masks (shown as grey in Figure 19), which approximately described the locations of the cirrus cloud at the
times of CI. Observations of the cirrus cloud made by
the 3 GHz radar at Chilbolton indicated that the base
of the cirrus was located between 5.5 km and 6.5 km.
Radiosonde data showed that the temperature at 6 km
was around 250 K, so a BT threshold of 250 K was used
Copyright  2007 Royal Meteorological Society
to define the area covered by the cirrus clouds. This simple threshold used to define the cirrus mask is fairly crude
(for example, it does not take account of thin cirrus that
appears warmer than this threshold), but it is useful as
a means of classifying the location of initiation of the
showers in terms of whether they occur under cloudy
or clear sky. In addition, the 250 K threshold used was
the maximum BT for which the cirrus shading had clear
effects on the water-vapour mixing ratio observed at midlevels in the boundary layer (Marsham et al., 2007, figure
8).
Determining the location of CI for the clouds that
developed into showers was not always easy, because
of obscuration by cirrus for certain periods of time, and
so the procedure is explained in some detail. For each
of the 25 showers, the times of first precipitation echo
(FPE) and first satellite cloud were determined from
the radar network and Meteosat high-resolution visible
data. The observed propagation speeds and directions of
the rain showers and cumulus clouds (when detected)
were then used to estimate a series of locations for the
developing convection in the hour preceding FPE, in case
this convection was hidden by cirrus cover during this
time. This was not a problem after 13:15 UTC, since after
this time all showers initiated in clear skies, so Figure 19
is limited to showing these tracks for showers that formed
before this time. If one of these back-trajectory locations
had been under the cirrus, then there would have been
the possibility that cumulus clouds formed earlier than
could be detected and were hidden by the cirrus; but this
was never observed to occur. Consequently, for showers
for which clouds were detectable in the satellite data,
the time of first observed convective cloud was taken as
the time of CI. In some cases, however, there were no
well-defined clouds in the visible imagery that could be
associated with the radar tracks. For these cases, the time
of CI was assumed to be the most recent time when the
back-trajectory was under clear sky (showers 4, 12, 13
and 19, with assumed CI 30–45 min before FPE). If this
never occurred (showers 11 and 17), then initiation must
have occurred under the cirrus, and the time of CI was
taken as 30 min before FPE (since 32 min was the mean
time difference between FPE and observed CI).
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
1100
(b)
100
80
80
60
60
40
40
20
16
0
-20
20
0
4
-20
-40
-40
-60
-60
-80
-80
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
(c)
1130
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
(d)
1145
100
100
80
80
60
60
40
20
40
13
9
0
-20
Distance (km)
Distance (km)
1115
100
Distance (km)
Distance (km)
(a)
20
0
-20
-40
-40
-60
-60
-80
-80
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
1657
1
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
Figure 19. Position of the cirrus mask (grey shading) when CI was estimated to have occurred. The tracks of the showers initiating at that time
are plotted using grey triangles and black lines (radar-derived). The tracks of the convective clouds (when seen by Meteosat) are plotted using
black circles and grey lines. Inferred locations from CI until a shower was observed are shown by asterisks. The coastline is shown by a black
line. Numbers are used to identify each shower. Initiation was observed to occur in clear skies after 13:15 UTC, and so these times are not
shown (showers 21 to 25). The figure is continued overleaf.
The above analysis showed 2 out of the 25 showers
initiating under the cirrus rather than in clear skies. The
mean fraction of sky occupied by the cirrus mask at the
times of CI was 14%, so a purely random set of CIs would
have had 3.5 ± 1.7 of the 25 showers initiating under the
cirrus (1.7 is the standard deviation of the binomial distribution). The observed value of 2 out of 25 is therefore
approximately one standard deviation below the mean.
Repeating this analysis, but excluding showers forming
after 13:15 UTC (since these formed in clear skies) gives
an observed value of 2 out of 25 showers, and an expected
value of 2.9 ± 1.6 for randomly-distributed showers.
These results only allow limited confidence in rejecting
the null hypothesis that the cirrus cover did not affect the
spatial distribution of the initiation of convection.
Closer inspection of Figure 19 shows that there appear
to be four main sub-regions of CI: first, at the leading
edge of the anvil (showers 3, 4, 8, 10, 12, 13 and 16);
Copyright  2007 Royal Meteorological Society
secondly, within or at the rear edge of the gap in the
cirrus that was situated east of Chilbolton at 12:00 UTC
(showers 17, 18, 19 and 20); thirdly, in clear skies behind
the trailing edge of the cirrus anvil (showers 2, 5, 6, 7,
14, 15 and 21); and finally, in clear skies far from the
cirrus (showers 1 and 9, as well as showers 22, 23, 24 and
25, which are not shown but initiated in clear skies after
13:15 UTC). If showers 4, 12, 13 and 19 are assumed to
initiate 30 min before FPE, rather than at their last clearsky location, these categorizations are unchanged. This
pattern of behaviour, with observed CI occurring close
to the rear edge of gaps, at the leading edge of the cirrus
anvil and in clear skies, is consistent with the modelling
studies described in Section 4, which show convergence
at the leading edge of the anvil and at the rear edge
of gaps. Shower 11 is the only clear exception to this
pattern, since it initiates under the cirrus anvil (shower
17 initiates under the cirrus, but at the rear edge of a gap).
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1658
J. H. MARSHAM ET AL.
1200
(e)
1230
100
(f)
80
80
12
60
3
20
0
-20
18
40
8
Distance (km)
Distance (km)
60
10
40
20
100
2
14
20
17
0
-20
-40
-40
-60
-60
-80
-80
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
1245
(g)
1300
(h)
100
80
80
11
60
60
19
40
Distance (km)
Distance (km)
40
20
0
100
15
-20
7
20
6
5
0
-20
-40
-40
-60
-60
-80
-80
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
-100
-100 -80 -60 -40 -20 0 20 40 60 80 100
Distance (km)
Figure 19. (Continued).
Finally, it is also interesting to note that 6 out of the
25 observed CI locations were within 5 km of the edge
of the cirrus mask. If showers were randomly located,
we would expect this to be 1.9 ± 1.3 out of 25 (on
average, 7.6% of the region was within 5 km of the
edge of the mask). Again, repeating this analysis for only
showers forming before 13:15 UTC gives an observed
value of 6 showers, and an expected value of 1.6 ±
1.2 for randomly-distributed showers. These observed
values are significantly larger than those expected from a
random distribution of showers, strongly supporting the
hypothesis that convergence at the edge of the cirrus anvil
affected CI.
Figure 20 shows inferred locations of CI in relation to
the orography of the CSIP area. It is clear that showers
initiated further south, where the hills also extended
further south, suggesting that orographic effects were
also significant. Figure 20 also shows that the showers
that were inferred to have initiated under the cirrus (11
and 17) formed on the south-facing (windward) slopes of
hills, suggesting that orographic effects may have been
significant in these cases.
Copyright  2007 Royal Meteorological Society
6.
Conclusions
The results described lead us to believe that shading from
cirrus anvils had significant effects on CI during CSIP
IOP 5. In LEM simulations, based on observations from
IOP 5, cirrus shading suppressed CI, and the rear edge of
gaps in the cirrus cover and the leading edge of the cirrus
anvil were preferred locations for initiation, because of
the convergence there. Observations of CI from rain-radar
and Meteosat satellite data support the hypothesis that
the leading edge of the anvil, the rear edge of gaps in
the cirrus and clear-sky regions were preferred locations
for initiation. These results are consistent with (Roebber
et al., 2002), but for CSIP IOP 5 the effects of the cirrus
shading on surface fluxes have been quantified, LEM
simulations have been used to examine the mechanisms,
and the observed distribution of initiation events has been
evaluated statistically. In addition, the effects of the cirrus
shading on the boundary layer, described in (Marsham
et al., 2007), provide further support for the hypotheses
that variations in cirrus cloud cover led to mesoscale
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1659
CIRRUS SHADING DURING CSIP IOP 5. I: INITIATION OF CONVECTION
300
100
80
250
60
11
40
12
200
24
19
20
0
22
18
10
8
21
25
17
9
6
5
16
150
20
23
4
3
−20
−40
13
7
14 15
2
100
1
−60
50
−80
−100
−100
−80
−60
−40
−20
0
0
20
40
60
80
100
Figure 20. Locations of CI with respect to the height of the orography (m) in the CSIP area. Circles indicates that CI was determined from the
appearance of cumulus clouds in clear sky. Asterisks indicate that the location of CI was determined by extrapolating the shower track to the
latest time when the track is in clear sky (i.e. initiation occurred in clear sky, but the cirrus then came over and hid the cumuli). Squares indicate
that CI occurred under the cirrus and the location had to be inferred.
circulations and that cirrus shading significantly affected
CI during CSIP IOP 5.
Observations from Chilbolton showed a nearly-linear
variation of surface sensible and latent fluxes with the
downwelling solar flux. The cirrus anvils reduced the
downwelling solar flux, and so reduced surface fluxes.
For thick cirrus, this reduced the surface-sensible-heat
flux from the 200 Wm−2 expected with clear sky to
approximately 0–50 Wm−2 (Section 3). The effects of
shading by the cirrus were represented in the LEM by
simply using a moving surface-flux anomaly. Simulations
using a radiative-transfer model coupled to the LEM
showed that it was reasonable to neglect the shading
by, and radiative heating of, convective clouds, since
including these effects had little impact on the modelled
initiation (Section 4).
Two-dimensional LEM simulations, using a moving
positive flux anomaly to represent the effects of a gap
in the cirrus cover, were performed. In these runs, the
maximum cloud-top height reached a given level slightly
sooner than was predicted using a simple model, where
the time required depended only on the maximum amount
of heat added to any column of the model. This difference increased with increasing cloud-top height, and
points to the role of the convergence induced by the
moving anomaly in initiating the simulated convection.
The pressure gradients and uplift induced by the moving
anomalies were shown to be reasonably well predicted by
the theory outlined in Section 4.2, which used a Boussinesq approximation and convergence in two dimensions.
In addition, three-dimensional runs, performed using a
coarser grid spacing, showed that CI occurred more
Copyright  2007 Royal Meteorological Society
quickly with three dimensions, and this is likely to be due
to increased convergence. The modelled boundary layer
was deepest at and behind the rear edge of the moving
positive flux anomaly, since the amount of heat added to
the boundary layer was greatest behind this edge and the
convergence occurred at this edge. As a result, CI tended
to occur towards the rear edge of the moving positive
flux anomaly (i.e. at the centre of the reduced pressure).
LEM simulations using moving cold surface-flux
anomalies to represent the effects of the overall cirrus
anvil showed that the cold anomalies suppressed convection in their path, but gave convergence at their leading
edges. This convergence was again reasonably well predicted by the theory described in Section 4.2. With a
mean wind, which allowed convective clouds to move
more in phase with the moving anomaly, deeper convection persisted at the leading edge of the anomaly. Thus,
the LEM results, using both the positive and the negative
surface-flux anomalies, show that we expect CI at the
rear edge of gaps in the cirrus cover and at the leading
edge of the cirrus anvil.
A detailed analysis of the locations of CI observed
during IOP 5 (Section 5) showed that significantly more
CI occurred near the edge of the cirrus than would be
expected by chance. This is consistent with convergence
effects at the edge of the cirrus cloud, inferred from LEM
simulations. The analysis of the observations did not
statistically prove or disprove the hypothesis that showers
were preferentially initiated in clear skies. However, 24
of the 25 showers observed initiated at the rear edge of
gaps in the cirrus, at the leading edge of the cirrus anvil,
or in clear skies; this is consistent with the LEM results.
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj
1660
J. H. MARSHAM ET AL.
Of the 25 showers, 23 initiated in relatively clear skies,
rather than under the cirrus. In both cases, the exceptions
initiated over the south-facing slopes of significant hills,
suggesting that in these cases orographic effects were
significant. In addition, the observed CI occurred further
south to the west of Chilbolton, where hills also extend
further south, suggesting that the CI may have been more
generally affected by the orography.
In summary, this CSIP case shows that for forecasting
convective storms using NWP models it is important
to predict or assimilate cloud cover and account for
its effects on solar and surface fluxes. At the stage
of boundary-layer warming when the CIN has become
small, very small perturbations can determine precisely
where and when CI will occur (e.g. Morcrette et al.,
2006; Marsham and Parker, 2006). In such cases, even
small gaps in cloud cover, of approximately 30 km
diameter, can significantly affect CI, despite moving
relative to the surface. NWP models cannot be expected
to predict the detailed evolution of upper-level clouds
at long lead times, especially for highly inhomogeneous
anvils from earlier convective storms. This provides a
significant limitation on the predictability of CI in cases
where such subtle forcings are significant.
Acknowledgements
We would like to thank EUMETSAT for the Meteosat
data used in this paper, and the Met Office for the
Nimrod rain-rate images and the surface-pressure analysis
chart used. Observations from Chilbolton were courtesy
of the Chilbolton Facility for Atmospheric and Radio
Research (CFARR) at the CCLRC-Chilbolton Observatory, distributed via the NERC British Atmospheric
Data Centre (BADC). We would also like to thank all
those involved in the CSIP field campaigns, and the two
anonymous reviewers whose comments have improved
the clarity of this paper. This project was funded by
the Natural Environment Research Council (NERC):
NER/O/S/2002/00971.
References
Andreas EL, Jordan RE, Makshtas AP. 2005. Parameterizing turbulent
exchange over sea ice: the ice station Weddell results. BoundaryLayer Meteorol. 114: 439–460.
Bailey MJ, Carpenter LR, Lowther LR, Passant CW. 1981. A mesoscale
forecast for 14 August 1975 – the Hampstead storm. Meteorol. Mag.
110: 147–161.
Bennett LJ, Browning KA, Blyth AM, Parker DJ, Clark PA. 2006.
A review of the initiation of precipitating convection in the United
Kingdom. Q. J. R. Meteorol. Soc. 132: 1001–1020.
Betts AK, Ball JH, Beljaars ACM, Miller MJ, Viterbo PA. 1996.
The land surface–atmosphere interaction: A review based on
observational and global modeling perspectives. J. Geophys. Res.
101: 7209–7225.
Browning K, Blyth A, Clark P, Corsmeier U, Morcrette C, Agnew J,
Bamber D, Barthlott C, Bennett L, Beswick K, Bitter M, Bozier K,
Brooks B, Collier C, Cook C, Davies F, Deny B, Engelhardt M,
Feuerle T, Forbes R, Gaffard C, Gray M, Hankers R, Hewison T,
Huckle R, Kalthoff N, Khodayar S, Kohler M, Kraut S, Kunz M,
Ladd D, Lenfant J, Marsham J, McGregor J, Nicol J, Norton E,
Parker D, Perry F, Ramatschi M, Ricketts H, Roberts N, Russell
Copyright  2007 Royal Meteorological Society
A, Schulz H, Slack E, Vaughan G, Waight J, Watson R, Webb A,
Wieser A, Zink K. 2006a. The Convective Storm Initiation Project.
Bull. Am. Meteorol. Soc. (to appear).
Browning K, Morcrette C, Blyth A, Bennett L, Clark P, Corsmeier U,
Agnew J, Barkwith A, Barthlott C, Behrendt A, Bennett A, Beswick
K, Bozier K, Brooks B, Chalmers N, Collier C, Cook C, Dacre
H, Davies F, Davies L, Davies O, Deny B, Devine G, Dixon M,
Engelhardt M, Fitch A, Forbes R, Gaffard C, Gallagher M, Goddard
J, Gray M, Hewison T, Huckle R, Illingworth A, Kalthoff N, Keeley
S, Khodayar S, Kilburn C, Kohler M, Kottmeier C, Kraut S, Ladd
D, Lean H, Lenfant J, Marsham J, McGregor J, Mobbs S, Nicol J,
Norton E, Parker D, Perry F, Ramatschi M, Richardson A, Ricketts
H, Roberts N, Russell A, Slack E, Vaughan G, Watson R, Webb A,
Weckwerth T, Wieser A, Wilson J, Wrench C, Wulfmeyer V, Zink
K. 2006b. ‘A summary of the Convective Storm Initiation Project
Intensive Observation Periods’. Met Office Forecasting Research
Technical Report: 474/JCMM Report 153.
Clark P, Lean H. 2006. ‘An overview of high resolution UM
performance for CSIP cases’. Met Office Forecasting Research
Technical Report: 478/JCMM Report 155.
Fu Q, Liou KN. 1992. On the correlated k-distribution method for
radiative transfer in nonhomogeneous atmospheres. J. Atmos. Sci.
49(22): 2139–2156.
Fu Q, Liou KN. 1993. Parameterization of the radiative properties of
cirrus clouds. J. Atmos. Sci. 50(13): 2008–2025.
Golding B, Clark P, May B. 2005. Boscastle flood: Meteorological
analysis of the conditions leading to the flooding on 16 August 2004.
Weather 60(8): 230–235.
Grabowski WW, Bechtold P, Cheng A, Forbes R, Hailliwell C,
Khairoutdinov M, Lang S, Nasuno T, Petch J, Tao WK, Wong R,
Wu X, Xu KM. 2006. Daytime convective development over land:
A model intercomparison based on LBA observations. Q. J. R.
Meteorol. Soc. 132: 317–344.
Gray MEB, Petch J, Derbyshire SH, Brown AR, Lock AP, Swann HA.
2001. ‘Version 2.3 of the Met. Office large eddy model’. Met Office
(APR) Turbulence and Diffusion Rep. 276. Met Office, Exeter, UK.
Johnson DB, Flament P, Bernstein RL. 1994. High-resolution satellite
imagery for mesoscale meteorological studies. Bull. Am. Meteorol.
Soc. 75: 5–33.
Kalthoff N, Fiebig-Wittmaack M, Meißner C, Kohler M, Uriarte M,
Bischoff-Gauß I, Gonzales E. 2006. The energy balance, evapotranspiration and nocturnal dew deposition of an arid valley in the
Andes. J. Arid Environ. 65: 420–443.
Marsham JH, Parker DJ. 2006. Secondary initiation of multiple bands
of cumulonimbus over southern Britain. II: Dynamics of secondary
initiation. Q. J. R. Meteorol. Soc. 132: 1053–1072.
Marsham JH, Blyth AM, Parker DJ, Beswick K, Browning KA,
Corsmeier U, Kalthoff N, Khodayar S, Morcertte CJ, Norton EG.
2007. Variable cirrus shading during CSIP IOP 5. II: Effects on the
convective boundary layer. Q. J. R. Meteorol. Soc. 133: 1661–1675.
McNider RT, Jedlovec JD, Wilson GS. 1984. Data analysis and
model simulation of the initiation of convection on April 24, 1982.
Pp. 543–549 in Tenth Conf. on Weather Forecasting and Analysis,
Tampa. American Meteorological Society.
Morcrette CJ, Browning KA, Blyth AM, Bozier KE, Clark PA, Ladd
D, Norton EG, Pavelin E. 2006. Secondary initiation of multiple
bands of cumulonimbus over southern Britain. I: An observational
case-study. Q. J. R. Meteorol. Soc. 132: 1021–1051.
Petch JC. 2006. Sensitivity studies of developing convection in a cloudresolving model. Q. J. R. Meteorol. Soc. 132: 345–358.
Roebber PJ, Scultz DM, Romero R. 2002. Synoptic regulation of the 3
May 1999 tornado outbreak. Weather and Forecasting 17: 399–429.
Segal M, Arritt RW. 1992. Nonclassical mesoscale circulations caused
by surface sensible heat-flux gradients. Bull. Am. Meteorol. Soc.
73(10): 1593–1604.
Segal M, Purdom JFW, Song JL, Pielke RA, Mahrer Y. 1986.
Evaluation of cloud shading effects on the generation and
modification of mesoscale circulations. Mon. Weather Rev. 114:
1201–1212.
Simpson PM, Brand EC, Wrench CL. 2002. ‘Microwave radiometer
measurements at Chilbolton – liquid water path algorithm development and accuracy’. EU FP5-CloudNET Project Report.
Thielen J, Wobrock W, Gadian A, Mestayer PG, Creutin JD. 2000.
The possible influence of urban surfaces on rainfall development:
a sensitivity study in 2D in the meso-γ -scale. Atmos. Res. 54(1):
15–39.
Q. J. R. Meteorol. Soc. 133: 1643–1660 (2007)
DOI: 10.1002/qj